An Advanced Group Contribution Method for High-Dimensional, Sparse Data Sets.

Area of Science:

• Chemical Engineering
• Computational Chemistry
• Physical Chemistry

Background:

• Accurate physical property data is crucial for chemical process design and operation.
• Literature data for all chemical components is often unavailable.
• Predictive methods like group contribution methods (GCMs) offer cost-effective alternatives to experimental measurements.

Purpose of the Study:

• To develop a novel group contribution method (GCM) for predicting chemical properties.
• To enhance the accuracy and applicability of GCMs for high-dimensional, sparse datasets.
• To compare the performance of the proposed GCM against existing methods.

Main Methods:

• Developed a new GCM by extending the database and classifying compounds into non-ring and ring groups.
• Employed Support Vector Regression (SVR) as the core regression model.
• Integrated Particle Swarm Optimization (PSO) for derivative-free parameter optimization of the SVR model to prevent local optima.

Main Results:

• The proposed GCM demonstrates improved applicability and accuracy for sparse data.
• The PSO-optimized SVR model effectively avoids local optimality in parameter tuning.
• Performance evaluation indicates competitive or superior results compared to other GCMs.

Conclusions:

• The novel GCM, enhanced by SVR and PSO, provides a robust approach for predicting chemical properties.
• This method is particularly effective for datasets characterized by high dimensionality and sparsity.
• The findings contribute to more efficient and accurate chemical process design and operation.